Browse the metahub catalog

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable evaluation on local Docker, Slurm HPC, or cloud platforms. NVIDIA's enterprise-grade platform with container-first architecture for reproducible benchmarking.

Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for fast quantization workflows, or when deploying with vLLM or HuggingFace Transformers.

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.

Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequences (>512 tokens), encountering GPU memory issues with attention, or need faster inference. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.

Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.

Serverless GPU cloud platform for running ML workloads. Use when you need on-demand GPU access without infrastructure management, deploying ML models as APIs, or running batch jobs with automatic scaling.

Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent filesystems, or high-performance multi-node clusters for large-scale training.

Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.

Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.

Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.

Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention

Meta's 86M prompt injection and jailbreak detector. Filters malicious prompts and third-party data for LLM apps. 99%+ TPR, <1% FPR. Fast (<2ms GPU). Multilingual (8 languages). Deploy with HuggingFace or batch processing for RAG security.

NVIDIA's runtime safety framework for LLM applications. Features jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection. Uses Colang 2.0 DSL for programmable rails. Production-ready, runs on T4 GPU.

Meta's 7-8B specialized moderation model for LLM input/output filtering. 6 safety categories - violence/hate, sexual content, weapons, substances, self-harm, criminal planning. 94-95% accuracy. Deploy with vLLM, HuggingFace, Sagemaker. Integrates with NeMo Guardrails.

Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL from AI Feedback). Use for safety alignment, reducing harmful outputs without human labels. Powers Claude's safety system.

Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.

Workflow 4: Submission rebuttal pipeline. Parses external reviews, enforces coverage and grounding, drafts a safe text-only rebuttal under venue limits, and manages follow-up rounds. Use when user says "rebuttal", "reply to reviewers", "ICML rebuttal", "OpenReview response", or wants to answer external reviews safely.

Manage GPU compute jobs on the Qizhi (启智) platform using qzcli — a kubectl-style CLI tool. Use when user says "qzcli", "启智平台", "submit job", "stop job", "查计算组", "avail", "list jobs", "batch submit", or needs to manage distributed training jobs on a Qizhi instance.

Writes rigorous mathematical proofs for ML/AI theory. Use when asked to prove a theorem, lemma, proposition, or corollary, fill in missing proof steps, formalize a proof sketch, 补全证明, 写证明, 证明某个命题, or determine whether a claimed proof can actually be completed under the stated assumptions.

Rigorous mathematical proof verification and fixing workflow. Reads a LaTeX proof, identifies gaps via cross-model review (external reviewer backend, ultra reasoning), fixes each gap with full derivations, re-reviews, and generates an audit report. Use when user says "检查证明", "verify proof", "proof check", "审证明", "check this proof", or wants rigorous mathematical verification of a theory paper.

Search patent databases and academic literature for prior art relevant to an invention. Use when user says "现有技术检索", "prior art search", "专利检索", "check patents", or wants to find relevant prior art.

Generate pixel art SVG illustrations for READMEs, docs, or slides. Use when user says "画像素图", "pixel art", "make an SVG illustration", "README hero image", or wants a cute visual.

Get an external patent examiner review of a patent application. Use when user says "专利审查", "patent review", "审查意见", "examiner review", or wants critical feedback on patent claims and specification.

Full patent drafting pipeline from invention description to jurisdiction-formatted filing documents. Supports CN (CNIPA), US (USPTO), EP (EPO). Supports invention patents and utility models. Use when user says "写专利", "patent pipeline", "专利申请", "draft patent", "写权利要求书", or wants to draft a complete patent application.

Assess patent novelty and non-obviousness against prior art. Use when user says "专利查新", "patent novelty", "可专利性评估", "patentability check", or wants to evaluate if an invention is patentable.

Workflow 3: Full paper writing pipeline that goes from a narrative report to a polished, submission-ready PDF. Use when user says "写论文全流程", "write paper pipeline", "从报告到PDF", "paper writing", or wants the complete paper generation workflow.

Draft LaTeX paper section by section from an outline. Use when user says "写论文", "write paper", "draft LaTeX", "开始写", or wants to generate LaTeX content from a paper plan.

End-to-end conference talk pipeline: paper → slide outline → Beamer + PPTX → per-page polish → assurance checks (claim / citation / anonymity) → final export and report. Default-good for academic conference talks (NeurIPS / ICML / ICLR / VALSE / 投稿 talks). Trigger phrases: "做 talk", "做 PPT 全流程", "talk pipeline", "end-to-end slides", "做演讲", "conference talk full workflow". Use when the user wants the complete talk artifact, not just a slide deck.

Generate conference presentation slides (beamer LaTeX → PDF + editable PPTX) from a compiled paper, with speaker notes and full talk script. Use when user says "做PPT", "做幻灯片", "make slides", "conference talk", "presentation slides", "生成slides", "写演讲稿", or wants beamer slides for a conference talk.

DEPRECATED — superseded by /paper-poster-html. Kept only as a redirect for muscle memory; do not use for new posters.

DEFAULT poster pipeline — build an academic conference poster (ICML/NeurIPS/ICLR/CVPR/...) as a single HTML/CSS file with measurement-driven hard gates, real paper figures, a two-hue design-token system, and print-ready PDF via headless Chromium. Use when the user says "做海报", "poster", "conference poster", "paper poster", or asks to design/redo a research poster. Supersedes the retired LaTeX /paper-poster.

Generate a structured paper outline from review conclusions and experiment results. Use when user says "写大纲", "paper outline", "plan the paper", "论文规划", or wants to create a paper plan before writing.

Generate publication-quality AI illustrations for academic papers using Gemini image generation. Creates architecture diagrams, method illustrations with Claude-supervised iterative refinement loop. Use when user says "生成图表", "画架构图", "AI绘图", "paper illustration", "generate diagram", or needs visual figures for papers.

Generate publication-quality academic illustrations through a local Codex app-server bridge that uses Codex native image generation. This is a separate experimental alternative to `paper-illustration`, intended for Claude Code users who want a GPT-image-style renderer without modifying the original skill.

Generate publication-quality figures and tables from experiment results. Use when user says "画图", "作图", "generate figures", "paper figures", or needs plots for a paper.

Compile LaTeX paper to PDF, fix errors, and verify output. Use when user says "编译论文", "compile paper", "build PDF", "生成PDF", or wants to compile LaTeX into a submission-ready PDF.

Zero-context verification that every number, comparison, and scope claim in the paper matches raw result files. Uses a fresh cross-model reviewer with NO prior context to prevent confirmation bias. Use when user says "审查论文数据", "check paper claims", "verify numbers", "论文数字核对", or before submission to ensure paper-to-evidence fidelity.

Two-way sync between a local paper directory and an Overleaf project, so ARIS audit/edit workflows stay on the local copy while collaborators edit in the Overleaf web UI. Use when user says "同步 overleaf", "overleaf sync", "推送到 overleaf", "connect overleaf", "Overleaf 桥接", "pull overleaf", "push overleaf", or wants to bridge their ARIS paper directory with an Overleaf project.

Search academic papers via OpenAlex API for open citation data, institutional affiliations, and funding information. Use when user says "openalex search", "search openalex", "open citation graph", or wants comprehensive academic metadata beyond arXiv/Semantic Scholar.

Verify research idea novelty against recent literature. Use when user says "查新", "novelty check", "有没有人做过", "check novelty", or wants to verify a research idea is novel before implementing.

Monitor running experiments, check progress, collect results. Use when user says "check results", "is it done", "monitor", or wants experiment output.

Analyze ARIS usage logs and propose optimizations to SKILL.md files, reviewer prompts, and workflow defaults. Outer-loop harness optimization inspired by Meta-Harness (Lee et al., 2026). Use when user says "优化技能", "meta optimize", "improve skills", "分析使用记录", or wants to optimize ARIS's own harness components based on accumulated experience.

Generate Mermaid diagrams from user requirements. Supports flowcharts, sequence diagrams, class diagrams, ER diagrams, Gantt charts, and 18 more diagram types.

Two-thread adversarial review: a fresh reviewer constructs the strongest 200-word rejection memo, then a second fresh reviewer defends the paper point-by-point and surfaces still-unresolved critical issues. Use when user says "kill argument", "adversarial review", "hostile review", "rebuttal preparation", "reviewer-2 simulation", or before submitting a theory paper that has already passed standard review rounds.

Compile patent application into jurisdiction-specific filing format. Use when user says "格式转换", "jurisdiction format", "国家格式", "compile patent", or wants formatted patent documents for CN/US/EP filing.